>Oh - and I understand it's best to use NEWSEQUENTIALID() for clustering

Or you can do like me and use non clustered indexes for the GUID PK:)

You could use COMB GUID for clustered indexes.
Check Jimmy Nilson's benchmarks (it shows insertion uisng GUID is 10% slower only):

http://www.informit.com/articles/article.aspx?p=25862

Cetin